Fix parse_csv() dropping the final CSV field

[mikeysklar]

Summary

parse_csv() in src/AdafruitIO_Data.cpp silently drops the final field of every CSV it parses — for an N-field input it only populates slots 0..N-2, leaving NULL at N-1.

Root cause

The '\0' case in the parse loop breaks out without emitting the field accumulated since the last comma, unlike the ',' case which correctly strdups each field.

Fix

Emit the final field from inside the '\0' case before setting fEnd, using the same strdup + cleanup path as the ',' case.

Relationship to #180

#180's index fix in _parseCSV() is correct but depends on this PR — without it, correcting the indices causes atof(NULL) → crash on every incoming location message. Suggest landing this first, then #180.

Verification

Tested on Feather ESP32 V2 with both fixes applied:

=== parse_csv verification ===
lat: 10.111111
lon: 20.222222
ele: 30.333333

Test plan

• Unit-style sketch on Feather ESP32 V2: all four fields parsed correctly, no crash.
• - [ ] Maintainer sanity-check on another platform if desired.
  🤖 Generated with Claude Code

[mikeysklar]

End-to-end verified on Feather ESP32 V2 and QT Py ESP32-S3

Ran the stock examples/adafruitio_04_location sketch with both this PR and #180 applied on both an Adafruit Feather ESP32 V2 and an Adafruit QT Py ESP32-S3. Consecutive round-trips through the real AIO broker, every field matching, value incrementing across cycles (no more reboot loop), no crash on either board.

Feather ESP32 V2:

----- sending -----
value: 1
lat: 42.321427
lon: -83.025754
ele: 1.00
----- received -----
value: 1
lat: 42.321427
lon: -83.025754
ele: 1.00
----- sending -----
value: 2
lat: 42.311427
lon: -83.005754
ele: 2.00
----- received -----
value: 2
lat: 42.311427
lon: -83.005754
ele: 2.00
----- sending -----
value: 3
lat: 42.301427
lon: -82.985754
ele: 3.00
----- received -----
value: 3
lat: 42.301427
lon: -82.985754
ele: 3.00
----- sending -----
value: 4
lat: 42.291427
lon: -82.965754
ele: 4.00
----- received -----
value: 4
lat: 42.291427
lon: -82.965754
ele: 4.00

Note: on main the S3 was crashing in what looked like the publish path (right after the ele: 0.00 print), but it was actually the same atof(NULL) deref hit via the inbound echo on a tighter timing window — the save() call returns, the MQTT thread delivers the echoed message a few ms later, and _parseCSV() crashes before loop() gets back to its next print. The combined #180 + #181 patches fix both boards.

The two fixes

• Fix parse_csv() dropping the final CSV field #181 (this PR) — parse_csv() stopped dropping the final CSV field. One-char cause: the '\0' case never emitted the accumulated tail into buf, so an N-field CSV came back with NULL in slot N-1.
• Fix _parseCSV() array index bug: lon and ele always returned lat value #180 — _parseCSV() was reading fields[1] three times for lat/lon/ele instead of fields[1]/[2]/[3].

They mask each other: #180's index bug kept #180's readers away from fields[N-1], hiding #181's NULL. Fixing #180 alone turns "wrong values" into "null-pointer deref on every receive" (LoadProhibited on ESP32). Both are needed; this one should land first.